Available online 28 October 2022, 100546

Few studies of environment effects on BMI account for spatial autocorrelation.

Spatial matrices based on proximity between participant addresses were compared.

Evidence of positive spatial autocorrelation in BMI in Adelaide but not Melbourne.

Mean BMI was higher in non-20-minute neighbourhoods than 20-minute neighbourhoods.

Modelling results were similar irrespective of spatial autocorrelation account.

Beliefs that neighbourhood environments influence body mass index (BMI) assume people residing proximally have similar outcomes. However, spatial relationships are rarely examined. We considered spatial autocorrelation when estimating associations between neighbourhood environments and BMI in two Australian cities. Using cross-sectional data from 1329 participants (Melbourne=637, Adelaide=692), spatial autocorrelation in BMI was examined for different spatial weights definitions. Spatial and ordinary least squares regression were compared to assess how accounting for spatial autocorrelation influenced model findings. Geocoded household addresses were used to generate matrices based on distances between addresses. We found low positive spatial autocorrelation in BMI; magnitudes differed by matrix choice, highlighting the need for careful consideration of appropriate spatial weighting. Results indicated statistical evidence of spatial autocorrelation in Adelaide but not Melbourne. Model findings were comparable, with no residual spatial autocorrelation after adjustment for confounders. Future neighbourhoods and BMI research should examine spatial autocorrelation, accounting for this where necessary.

Built environment

urban planning

health-environment association

spatial analysis

spatial autocorrelation

The authors do not have permission to share data.

© 2022 Published by Elsevier Ltd.